Oleksandr Levchenko: Financing Ukraine’s Recovery Must Combine Capital, Risk Management, and Responsible AI

Oleksandr Levchenko is a Candidate of Economic Sciences and a certified bank auditor who previously worked in banking inspection at the National Bank of Ukraine (NBU). Since January 2026, he has served as a Lead Research Fellow (on a voluntary basis) at S.A. Podolynsky Institute of Physical Economics, Private Enterprise located in Kyiv, Ukraine.

He has authored two copyright-registered methodologies and co-authored published research on explainable artificial intelligence (AI), regulatory compliance, and virtual assets. His work connects recovery financing with risk management and digital resilience.

Iryna Skoryk: What did banking inspection teach you about financing Ukraine’s recovery?

Oleksandr Levchenko: Banking inspection and audit taught me that a compliant report is not proof of a resilient institution. Economic cybernetics shaped my analytical approach. Studying financial services at Napier University in Edinburgh added practical context. On-site inspections, including information-security reviews, reinforced the need for verifiable data and an audit trail. Recovery financing needs that discipline: investors must know how risks are assessed and controlled, using comparable data and predictable rules.

Which of your dissertation findings have been used in practice?

I developed a method for assessing the efficiency of leasing transactions from the lessor’s perspective using an integrated risk indicator. It incorporates risk, uncertainty, and market instability into the assessment rather than adding them as a note after calculating returns.

Other Topics of Interest

‘Putin Is Using Hitler’s Tactics’ – Ranking US Senator Slams Russian Attacks on Ukrainian Cities

US lawmakers also ridicule Witkoff’s description of Putin as a “humanitarian,” pointing to Russia’s killing of civilians and attacks on energy infrastructure.

ULF-FINANS LLC applied that approach. The Ukrainian Union of Lessors used elements of my proposed financial and credit framework, while Banking University used results in teaching and research methodology. These documents establish practical use, not a measured increase in profit or reduction in losses. Nor does the use of those earlier results establish implementation of my newer methodologies.

What do PRISM-Lease and DFR-PRISM add, and what remains to be tested?

I developed both methodologies to address fragmented assessment and registered copyright in the works this year. PRISM-Lease integrates prudential compliance, risk management, and internal control in leasing companies.

DFR-PRISM addresses digital financial resilience and cyber risks in non-bank financial institutions, linking service disruptions, data risks, and provider dependencies to financial and governance consequences.

The aim is to make interconnected risks visible and support a timely management response, rather than examine each indicator in isolation. Copyright registration establishes authorship, not effectiveness – neither is an implemented product, and both still require independent validation and pilot testing.

Testing must also establish whether specialists understand the signals and can act on them.

What did your Aug. 31 compliance study show?

In our article published Aug. 31, my co-authors and I compared public NBU data with regulatory requirements. Our approach connects financial capacity, digital controls, legal obligations, governance, and professional competence.

Digital tools support reliable data and monitoring – legal mechanisms translate requirements into internal responsibilities.

Common control principles are useful, but the indicators, depth of review, and response procedures must fit the business model. An insurer, finance company, and credit union cannot be assessed adequately through identical checklists.

What did you contribute to the US auto lease study, and what did the tests show?

Our collaborative study of explainable AI, published in August, used public US Securities and Exchange Commission (SEC) data on six families of auto lease securitizations.

My contribution covered methodology, data preparation and curation, software implementation, formal analysis, visualization, and project administration.

We examined transitions to at least 30 days past due in the following month, not lifetime default. We separated time periods, tested contracts excluded from training, and assessed transfer to previously unseen issuers.

At equal review capacity, X-LEASE did not outperform the unconstrained model in detecting delinquency in our test. We analyzed explanations for both models: a more concentrated explanation did not establish greater fairness or auditability.

The findings support early-warning monitoring with human review under the conditions studied, not automatic adverse decisions or claims of universal transferability.

How would regulatory whitelists reduce wrongful payment blocks without weakening safeguards?

In our joint publication, we proposed validating the legal basis, recipient, and payment details of mandatory payments before blocking them.

This is an additional verification layer, not a privilege for recipients or an exemption from sanctions and anti-money-laundering controls. Conflicting signals require authorized human review.

The framework remains conceptual: it aims to reduce false-positive blocks without missing more dangerous transactions, but we have not tested its effectiveness on proprietary bank transaction data.

What does your Conceptual Prudential Compatibility Index (CPCI) establish – and what does it not?

In our joint work, CPCI distinguishes an asset’s tradability from its suitability to back an insurer’s or pension fund’s obligations. It assigns categories using documentary evidence on liquidity, value stability, and valuation reliability, then separately examines regulatory admissibility for the instrument, jurisdiction, institution, and intended use.

Legal prohibitions cannot be offset by attractive returns. CPCI helps identify gaps in the evidence – it is an ordinal framework, not an externally validated quantitative risk model, a price forecast, or a substitute for applicable law.

What did you propose to the NBU, and what would you pilot next?

I submitted my proposals on responsible AI to the NBU on July 13. They called for limited agent permissions, auditable actions, independent review proportionate to risk, and a tested safe-shutdown procedure. Customers must also have a genuine path to review and correction.

The NBU acknowledged receipt that day – this was not approval of the recommendations.

I would next seek independent pilot testing of an integrated early-warning system combining prudential, digital, and model risks, operating alongside existing procedures without automatically affecting customer decisions.